An Optimized Deep-Learning-Based Low Power Approximate Multiplier Design

Approximate computing is a popular field for low power consumption that used in several applications like image processing, video multimedia and data mining. This majorly performed with an arithmetic circuit particular multiplier. The multiplier the most essential element approximate where based on its performance. There are researchers worked reduction few decades, but design of not so easy. seems bigger challenge digital industries to minimum error rate higher accuracy. To overcome these issues, circuits applied Deep Learning (DL) approaches In recent times, DL method learning prediction accuracy fields. Therefore, Long Short-Term Memory (LSTM) time series this work computing. provide optimal solution, LSTM combined meta-heuristics Jellyfish search optimisation technique input aware deep learning-based (DLAM). work, jelly optimised model enhance metrics performance hyperparameters identified by optimisation. fine-tuning obtain solution perform proposed pre-trained generate libraries different truncation levels as function area, delay, metrics. experimental results 8-bit processing application shows achieved superior area very good rates.